RELIABILITY-CENTERED MAINTENANCE OPTIMIZATION USING MULTI-OBJECTIVE AI ALGORITHMS IN REFINERY EQUIPMENT

Abstract

This quantitative study on Reliability-Centered Maintenance Optimization Using Multi-Objective AI Algorithms in Refinery Equipment had examined how reliability-centered maintenance (RCM) logic could be translated into measurable decision variables and optimized under refinery constraints using multi-objective AI. A structured review of 62 peer-reviewed papers had informed objective selection, constraint definition, and algorithm benchmarking practices. The empirical analysis had used a retrospective refinery case dataset comprising 420 assets observed over 60 months (190 rotating, 170 static, 60 instrumentation/control) with 17,980,000 operating exposure hours. The dataset had included 612 functional failure events, 9,284 work orders (59.0% preventive, 41.0% corrective), and 4,116 inspection observations. Downtime severity had been right-skewed, with median downtime per event of 6.2 hours (IQR 2.1–18.4) and a 95th percentile of 92.0 hours, with static equipment exhibiting longer restoration burdens. Composite indices for operating severity, monitoring coverage, execution burden, degradation evidence, and risk exposure had demonstrated acceptable internal consistency (0.77–0.85). Collinearity screening reduced model predictors and improved stability, lowering maximum VIF from 9.4 to 2.8 in rotating equipment models. Regression results showed that operating severity increased functional failure hazard (HR 1.29, 95% CI 1.18–1.41, p < 0.001) and degradation evidence increased hazard (HR 1.33, 95% CI 1.22–1.46, p < 0.001), while monitoring coverage reduced hazard (HR 0.83, 95% CI 0.75–0.92, p = 0.001). Preventive intensity reduced corrective recurrence (IRR 0.79, 95% CI 0.70–0.90, p < 0.001) but increased direct cost (+5.4%, p = 0.001). Turnaround proximity reduced unavailability (RR 0.86, p = 0.001). Safety-critical assets showed higher inspection adherence (96.1% vs 88.7%; OR 2.72, p < 0.001). The findings had clarified the trade-off structure motivating Pareto-efficient maintenance portfolios that remained interpretable and constraint-feasible in refinery operations.